]> git.bitcoin.ninja Git - rust-lightning/commitdiff
Drop unnecessary SIMD subtraction in ChaCha20 `round`
authorMatt Corallo <git@bluematt.me>
Thu, 30 Nov 2023 23:35:43 +0000 (23:35 +0000)
committerMatt Corallo <git@bluematt.me>
Thu, 30 Nov 2023 23:35:43 +0000 (23:35 +0000)
While its all constant arithmetic to calculate the shift, which
LLVM likely optimizes out for us, there's no reason to do it four
times, which just makes the code harder to read.

lightning/src/util/chacha20.rs

index f46b344f2ce144c235e40e041743e707570e43a5..87500d88d5206c67d3a4d93a1d212a5a018e29cb 100644 (file)
@@ -43,16 +43,16 @@ mod real_chacha {
                        u32x4(self.0 ^ rhs.0, self.1 ^ rhs.1, self.2 ^ rhs.2, self.3 ^ rhs.3)
                }
        }
-       impl ::core::ops::Shr<u32x4> for u32x4 {
+       impl ::core::ops::Shr<u8> for u32x4 {
                type Output = u32x4;
-               fn shr(self, rhs: u32x4) -> u32x4 {
-                       u32x4(self.0 >> rhs.0, self.1 >> rhs.1, self.2 >> rhs.2, self.3 >> rhs.3)
+               fn shr(self, shr: u8) -> u32x4 {
+                       u32x4(self.0 >> shr, self.1 >> shr, self.2 >> shr, self.3 >> shr)
                }
        }
-       impl ::core::ops::Shl<u32x4> for u32x4 {
+       impl ::core::ops::Shl<u8> for u32x4 {
                type Output = u32x4;
-               fn shl(self, rhs: u32x4) -> u32x4 {
-                       u32x4(self.0 << rhs.0, self.1 << rhs.1, self.2 << rhs.2, self.3 << rhs.3)
+               fn shl(self, shl: u8) -> u32x4 {
+                       u32x4(self.0 << shl, self.1 << shl, self.2 << shl, self.3 << shl)
                }
        }
        impl u32x4 {
@@ -118,31 +118,25 @@ mod real_chacha {
        macro_rules! round{
                ($state: expr) => {{
                        $state.a = $state.a + $state.b;
-                       rotate!($state.d, $state.a, S16);
+                       rotate!($state.d, $state.a, 16);
                        $state.c = $state.c + $state.d;
-                       rotate!($state.b, $state.c, S12);
+                       rotate!($state.b, $state.c, 12);
                        $state.a = $state.a + $state.b;
-                       rotate!($state.d, $state.a, S8);
+                       rotate!($state.d, $state.a, 8);
                        $state.c = $state.c + $state.d;
-                       rotate!($state.b, $state.c, S7);
+                       rotate!($state.b, $state.c, 7);
                }}
        }
 
        macro_rules! rotate {
-               ($a: expr, $b: expr, $c:expr) => {{
+               ($a: expr, $b: expr, $rot: expr) => {{
                        let v = $a ^ $b;
-                       let r = S32 - $c;
+                       let r = 32 - $rot;
                        let right = v >> r;
-                       $a = (v << $c) ^ right
+                       $a = (v << $rot) ^ right
                }}
        }
 
-       const S32:u32x4 = u32x4(32, 32, 32, 32);
-       const S16:u32x4 = u32x4(16, 16, 16, 16);
-       const S12:u32x4 = u32x4(12, 12, 12, 12);
-       const S8:u32x4 = u32x4(8, 8, 8, 8);
-       const S7:u32x4 = u32x4(7, 7, 7, 7);
-
        impl ChaCha20 {
                pub fn new(key: &[u8], nonce: &[u8]) -> ChaCha20 {
                        assert!(key.len() == 16 || key.len() == 32);